Evaluating the Impacts of Eliminating Prosecutorial ......who run on a platform of change.1 One...

Evaluating the Impacts of Eliminating ProsecutorialRequests for Cash Bail

Aurelie Ouss∗and Megan Stevenson†‡

February 17, 2019

Abstract

Recent criminal justice reform efforts have focused on electing progressive prose-cutors to implement change, such as the reduction of cash bail as a requirement forpretrial release. However, critics worry that removing cash bail will decrease account-ability and increase failure-to-appear in court. We test this by looking at the effectsof the No-Cash-Bail reform policy initiated by Philadelphia’s recently elected DistrictAttorney, Larry Krasner. Under this policy, the DA’s office stopped requesting cashbail for defendants charged with a large variety of different offenses, both misdemeanorand felony. This policy led to an immediate 23% increase (12 percentage points) in thefraction of eligible defendants released with no monetary or other conditions (ROR),and a 22% (5 percentage points) decrease in the fraction of defendants who spent atleast one night in jail, but no detectable difference for longer jail stays. The maineffect of this policy was therefore to reduce the use of collateral to incentivize courtappearance. In spite of this large decrease in the fraction of defendants having mone-tary incentives to show up to court, we detect no change in failure-to-appear in courtor in recidivism, suggesting that reductions in the use of monetary bail can be madewithout significant adverse consequences. These results also demonstrate the role ofprosecutors in determining outcomes over which they have no direct authority, such assetting bail.

∗Assistant Professor of Criminology, University of Pennsylvania, [email protected]†Assistant Professor of Law, George Mason University, [email protected]‡Many thanks to the various individuals who provided help in this research, including Jennifer Doleac,

Oren Gur, Michael Hollander, Mark Houldin, Jacob Kaplan, John MacDonald, Alex Malek, Lyandra Retacco,Liam Riley, Ariel Shapell, and Benjamin Waxman.

Electronic copy available at: https://ssrn.com/abstract=3335138

2

1 Introduction

Prosecutors are thought to hold a substantial amount of power within the criminal

justice system. They make important discretionary decisions throughout the proceed-

ings: they choose the initial charge, they make bail requests, they can add or subtract

charges that carry mandatory minimums, they make evidence more or less accessible

for discovery, they can offer or reject plea deals, and so on. Approximately 95 percent

of convictions in federal and state courts are resolved by a defendant pleading guilty

(Reaves, 2013), so juries seldom act as a check on prosecutors. Criminal justice reform

efforts have recently mobilized to elect “progressive” prosecutors: district attorneys

who run on a platform of change.1 One common reform initiative championed by

progressive prosecutors is the reduction or elimination of cash bail as a condition of

pretrial release. Reformers argue that cash bail entails discrimination on the basis of

wealth, and results in low-risk defendants sitting in jail due to poverty. Given that

pretrial detainees account for 1/5 of the entire incarcerated population in the United

States and the majority of jail populations (Walmsley, 2014), such reforms could have

substantial impact.

Eliminating cash bail, however, could come with unwanted side effects. The cash

deposit is designed to incentivize appearance in court. With no “skin in the game”

defendants may fail to appear in court, thus imposing costs on both the courts and the

police who are tasked with retrieving them. Furthermore, critics worry that releasing

more arrestees without imposing more supervision or other alternatives to incarceration

could result in an increase in pretrial crime. Several recent papers have shown long-

term negative effects of pretrial detention on defendant outcomes (Gupta et al., 2016;

Heaton et al., 2017; Leslie and Pope, 2017; Dobbie et al., 2018; Stevenson, 2018b), and

racial disparities in bail amounts and ability to post bail (Stevenson, 2018a). However,

there is little empirical work on the possible negative consequences of rolling back

pretrial detention and monetary bail.

In this paper, we present some of the first causal evidence of the effect of reducing

reliance on cash bail on court compliance and recidivism. We focus on the No-Cash-Bail

policy implemented by Philadelphia’s recently elected “progressive” District Attorney,

Larry Krasner. On February 21st, 2018, DA Krasner declared a new policy under

which the DA’s office would no longer seek cash bail for defendants charged with a

wide range of offense categories, both misdemeanor and felony. While DA Krasner

was elected on a liberal agenda, the exact nature and timing of his different reforms

was unknown. We argue that the announcement of the No-Cash-Bail reform created

a random shock in the bail process; in other words, that the only relevant difference

1See e.g. Larry Krasner (Philadelphia), Kim Ogg (Houston), Kim Fox (Chicago), Rachael Rollins(Boston), Marilyn Mosby (Baltimore).


3

between cases initiated immediately before versus immediately after was the change

in the DA’s bail policy. We use regression discontinuity to evaluate sharp changes to

pretrial outcomes that occurred at the time the No-Cash-Bail policy was implemented,

and a difference-in-differences approach to compare pretrial outcomes across case types

that were/were not eligible for the reform.

We first explore whether this policy changed bail amounts and detention rates. We

find that it resulted in an immediate 23% (12 percentage point) increase in the fraction

of eligible defendants who are released on their own recognizance (ROR), or released

without any monetary bail or other supervisory conditions. This is notable, since bail

decisions are made by magistrates who work for the judiciary, not for the prosecutor’s

office. It is not immediately apparent that a unilateral change initiated by the DA

would affect the magistrates’ bail decisions at all. We find that the new policy led to a

5 percentage point decline in the fraction of eligible defendants who spend at least one

night in jail but no statistically significant change in the fraction who spend a week or

more in jail. The policy had less of an impact on detention rates than ROR because

some of those who were released on recognizance as a result of the new policy would

have otherwise been released after the payment of monetary bail, or on unsecured or

supervised bail conditions. We find that most of the change comes from a decline in

the use of monetary bail in amounts of $5000 or less: a 7 percentage point, or 41%

relative decrease.2

Our results suggest that the main policy impact of the No-Cash-Bail reform was to

release more defendants without monetary conditions, with modest effects on pretrial

detention. This isolated effect, in turn, gives us an opportunity to determine how

monetary bail impacts court appearances and offending. We look at changes in several

outcomes during the four months after the initial bail hearing: failing to appear court

(FTA); being charged for a new offense; and being charged for a new serious offense. We

find no evidence that any of these outcomes got worse when defendants no longer had

monetary bail or other conditions of release. These results demonstrate that monetary

bail can be replaced by release on recognizance for a sizeable number of defendants

with no detectable impacts on non-appearance or pretrial crime.

This paper is one of the first to directly assess the effects of cash bail on court com-

pliance and offending, and to evaluate a “no cash bail” reform initiated by a prosecutor.

Given the many other jurisdictions that have enacted bail reform agendas including a

reduction or elimination of monetary bail,3 expanded research on the potential adverse

effects of such reforms is timely. Closest to our work is the Helland and Tabarrok

(2004) evaluation of cash bail, which uses propensity score matching. They find that

2Defendants in Philadelphia typically have to post 10% of the bail amount, so this would mean thatdefendants have to post $500 or less to get out of jail.

3New Jersey has eliminated cash bail almost entirely and California plans to eliminate it in October 2019.


4

felony defendants released with surety bonds are less likely to miss court appearances

than similar defendants released on recognizance, but these results may be partially

due to unobservable differences across defendants that weren’t accounted for in the

matching. Our paper also presents some of the first evidence on the role “progressive

prosecutors” can play in bail reform. Despite having no direct control over the bail

decision, we find that prosecutors still exert meaningful influence.

The remainder of the paper is organized as follows. In Section 2, we provide some

background on the use of cash bail in the United States, and describe the Philadelphia

No-Cash-Bail policy. In Section 3, we present our data and empirical strategy. Section

4 presents our main results, and Section 5 concludes.

2 Cash bail: General Background and Practices

in Philadelphia

2.1 Cash bail in the United States

Although criminal jurisdictions across the United States vary in the particulars of

the pretrial process, there are many commonalities.4. After arrest, a defendant is

brought to a police station where they are booked, and where they must remain until

a judge (or magistrate) has had an opportunity to review their case. At this initial

appearance, the judge will determine what conditions of release, if any, are appropriate

for the defendant. The judge may take input on this decision from the prosecutor’s

office or defense counsel, but the final decision is made by the judiciary. Conditions

of release may include the payment of monetary bail, an agreement to follow certain

conditions set by the court (e.g. curfew, drug testing, check-ins with a pretrial officer,

etc.), an agreement to pay a certain amount should the defendant fail to appear in

court (unsecured bond), and the promise of a third party (either a professional bail

bondsmen or a friend/family member) to supervise and ensure appearance. If the

defendant pays the monetary bail amount and/or agrees to the conditions they will be

released. If not, they will remain detained pretrial until the disposition of the case.

Monetary bail has long been one of the most dominant conditions of release. The

main goal of monetary bail is to ensure that those who are released from jail show up

in court for their appointed dates. Monetary bail acts as collateral; if the defendant

fails to appear in court, the bail amount will be forfeited. If a defendant cannot afford

their bail bond they may borrow it from a professional bail bondsmen. Monetary

bail is not supposed to be used either as a de-facto detention order or as a method

of preventing pretrial crime. Nonetheless, setting bail amounts at levels that prohibit

4For an overview see (Liu et al., 2018).


5

release is common. Only about 50% of defendants with monetary bail set are able to

pay to secure release, and, among jail inmates awaiting trial, 9/10 are detained on a

monetary bond (Cohen and Reaves, 2007).

Critics of the use of monetary bail argue that it is inefficient and can result in race

and class based disparities. The inefficiency argument stems from the fact that those

who cannot afford to pay will be detained even if they pose a relatively low risk of

non-appearance or reoffending, while defendants with greater financial resources may

be able to pay for release even if they pose a higher risk. This results in an ineffective

use of criminal justice resources. Furthermore, since race and wealth are correlated in

the United States, those who are detained solely due to inability to pay bail are likely

to be disproportionately people of color. This exacerbates racial disparities in pretrial

detention, and, due to the downstream consequences of detention, may exacerbate

racial disparities in conviction, sentencing, and future employment.

These concerns have led to a broad based movement to reform bail systems. Dozens

of jurisdictions across the United States have implemented, or will soon implement,

significant changes to their system of determining pretrial custody and ensuring court

appearance. Reducing or eliminating the use of monetary bail is a key component of

reform. In 2016, New Jersey essentially eliminated the use of monetary bail. California

is poised to do the same in 2019, after the passage of SB10. “Progressive” prosecutors

around the country have run on platforms of reducing the use of cash bail, and such

policies have already been implemented in many jurisdictions. However, bail reform

skeptics have expressed concern that reducing monetary bail could result – or already

has resulted – in unwanted consequences such as increasing non-appearance5 or crime.6

These concerns have slowed reform efforts in various jurisdictions, and threatens to

reverse some reforms that have already been implemented.7 Thus evidence on how

monetary bail impacts appearance rates and pretrial crime is crucially important to

the future of bail reform. In this paper, we evaluate the efficacy of monetary bail

as a mechanism to improve the functioning of courts: do monetary incentives have a

meaningful effect on compliance?

5E.g. according to some reports, non-appearance increased after Harris Countystopped detaining misdemeanor pretrial defendants who could not afford monetary bail.http://texasallianceforsafecommunities.org/victims/press-release-tasc-issues-statement-on-rising-failure-to-appear-rates-in-harris-county/

6E.g. DA Krasner’s bail reform policies have been blamed for increasing homicide rates in Philadelphia.https://6abc.com/head-of-philly-homicide-unit-reassigned-amid-growing-murder-rate/4976017/

7E.g. New Jersey’s Attorney General has started calling for pretrial detention in an expanded number ofcases. https://www.nj.com/politics/index.ssf/2017/05/nj supreme court tightens bail reform rules for gu.html


6

2.2 The pretrial process in Philadelphia

In Philadelphia, when a person is arrested, they are first brought to a local police sta-

tion where they are booked and placed into a holding cell. The police report associated

with the arrest will be sent to the DA’s office, where a prosecutor will review the case

and determine what, if any, charges to file. Once charges have been filed, the defendant

will be interviewed by Pretrial Services. Pretrial Services will make a recommendation

for the bail amount taking into account the defendants’ charges, criminal history and

life circumstances. Their recommendation is not binding and bail decisions are often

different (Shubik-Richards and Stemen, 2010). After the pretrial interview, the defen-

dant is ready for the bail hearing. The defendant remains in holding cell at the police

station and the bail hearing is held over videoconference. The bail hearing is presided

over by a magistrate who is an employee of the judiciary, and representatives of both

the DA’s office and the public defender’s office are present. While the representatives

can make suggestions for the appropriate bail amount, the final decision is made by the

magistrate. Neither the magistrate nor these representatives are, in general, attorneys.

The bail hearing typically lasts only a minute or two, during which the magistrate

reads the charges, schedules the next court date, determines eligibility for public de-

fense, and decides the conditions of release. These conditions include ROR (in which

the defendant is released solely on their promise to return to court), bail denied (de-

fendant in jail pretrial), monetary bail (defendant must provide financial collateral to

be released), unsecured bail (no need to post any money for release, but if the per-

son does not show up to their court date, they owe the court their bail amount), and

non-monetary bail (the defendant must agree to certain conditions, such as pretrial

supervision, but does not need to pay monetary bail). If the person fails to pay bail

within 4-8 hours of the bail hearing, they will be transported to the local jail. They will

remain there until the disposition of the case unless they can procure the bail deposit

or can convince a judge to lower it to an amount that is affordable.

Professional bail bondsmen are allowed in Philadelphia, but they are less common

than in other jurisdictions. This is because Philadelphia has a deposit system: the

defendant is released if they can pay 10% of the total bail amount. If they comply with

all release conditions 70% of the deposit will be returned when the case is disposed.

However if they fail to appear in court, they owe the court the full bail amount.

2.3 The Philadelphia No-Cash-Bail reform

On November 7th, 2017, Larry Krasner was elected to the position of Philadelphia’s

district attorney (DA). He ran on a progressive platform, which included goals like


7

lowering punishments for less serious crimes and reducing the use pretrial detention.8

However, and importantly for our research design, the exact timing of different reforms

was not announced ahead of time.

On February 21st, 2018, DA Krasner announced that his office would stop seeking

monetary bail if the lead charge was among a set of 25 low-level offenses.9 These offenses

include both felonies and misdemeanors, and spanned from very low-level offenses, such

as non-residential trespassing, to more severe offenses, such as burglaries with no person

present, and include several drug charges, such as possession with an intent to deliver.

The goal of this reform was to reduce pretrial detention, and to avoid incarcerating

defendants because they could not afford low bail amounts. Concretely, this meant

that the DA’s office would instruct their representatives at the bail hearing to ask that

defendants with these lead charges be released on their own recognizance, or to not

object if ROR was requested by the defendant’s legal representative.

This reform is not the only measure that DA Krasner took to reduce the reach of

criminal justice that winter. On Feb. 15th, the DA’s office announced a change in

charging practices for marijuana possession, retail theft and prostitution. Appendix

Figure A1 shows that after that date, the number of charges filed for these offenses

dropped. We remove them from our analyses. No other concurrent changes affected

the prosecution of low-level offenses, or pretrial detention.10

3 Data and empirical strategy

3.1 Data and descriptive statistics

Our data comes from the court records of the Pennsylvania Unified Judicial System.

We use web-scraped dockets for cases in which an initial bail hearing was held between

January 1st, 2017 and August 13th, 2018, in the Philadelphia Municipal Court. The

data contains one observation per criminal case. After dropping marijuana possession,

prostitution, and retail theft cases,11 as well as duplicate cases (i.e. a defendant is

brought for multiple cases on the same day),12 our data contains 47,052 observations.

8His agenda can be found here: https://krasnerforda.com/platform/9The full announcement and list of offenses eligible can be found here:

https://phillyda.wordpress.com/2018/02/21/larry-krasner-announces-end-to-cash-bail-in-philadelphia-for-low-level-offenses/

10Over the last several years, Philadelphia has introduced several other changes to their pretrial system,such as early bail review, in which a judge reviews bail for cases in which a defendant is unable to pay, and apilot project of providing pre-bail-hearing public defense to some defendants. However, these changes wereimplemented more than a year before the reforms evaluated in this paper and should not affect our analysis.

11Marijuana possession, retail theft and prostitution cases constitute ∼10% of pre-reform caseload.128.5% of defendants have multiple cases per day; we omit these due to difficulties in defining the bail type

for a defendant with multiple types of bail. Our results are very similar if we include duplicate cases.


8

Our dataset includes information on the defendant (first and last name, date of

birth, gender, race, address, and a unique court identifier), the offenses (date of arrest,

charges), the bail hearing (date and time of the bail hearing, type of attorney, initial

bail magistrate name, bail type and amount), whether and at what date and time

bail was posted, and notes pertaining to each court appearance (including whether the

defendant failed to appear). Using this data, we define several other main variables:

how much time, if any, a person spent in jail, and whether a person recidivated, recidi-

vated with a serious offense, or had an FTA within four months of their initial court

hearing.13 If the defendant had monetary bail, we define “time spent in jail” as the

time between when the initial bail was set and when bail was posted. Technically,

however, this includes both time spent in jail and time spent in the police holding cell,

since it can take a few hours between the time of the bail hearing and when defendants

are transferred to the jail. A person is considered to have spent “at least one night”

in jail if there is at least one calendar day difference between when bail was set and

posted; they are considered to have spent “at least two nights” in jail if they posted

bail two or more calendar days after their initial bail hearing; and to have spent “at

least a week” in jail if the bail posting date is at least 7 days after the initial bail

hearing.14 We generate a dummy for “recidivism” which is equal to one if a person

with the same first name, last name and date of birth is charged with a new offense

within four months of the bail hearing. Our “serious recidivism” variable is defined

similarly, except that we only count new charges for homicide or graded felonies of

type 1 – 3.15 Our FTA variable is equal to one if the defendant fails to appear for at

least one court date within four months of the bail hearing. Lastly, we define “eligible

cases” as cases that are eligible for the “No-Cash-Bail” policy; in other words, cases

for which the lead charge appears on the list of 25 offenses for which the DA’s office

would no longer request cash bail. Similarly, “ineligible cases” are cases whose lead

charge does not appear in that list of 25 offenses.

Table 1 presents descriptive statistics for cases filed before the “No Cash Bail”

reform was announced on February 21st. Statistics are shown separately for eligible

and ineligible cases. First, note that a majority of cases are eligible for no cash bail:

21,023 of the 33,358 cases in our sample, or 63%, are in the list of 25 offenses for

which the Philadelphia DA’s office was no longer going to seek monetary bail.16 This

13We look at a four month window in order to provide a consistent post-bail-hearing time window for alldefendants in our sample.

14Given that defendants already spent an average of 17 hours (Clark et al., 2011) in the holding cellawaiting the bail hearing, these definitions understate the true amount of time detained.

15This excludes, for instance, defendants who are charged only with possession-with-intent-to-deliver.16Overall, and including the case types omitted because of a concurrent change in charging practice,

approximately 67% of all cases filed in Philadelphia before the reform would have been eligible for theNo-Cash-Bail policy.


9

includes a significant number (43%) of felonies. Among eligible cases, 47% were already

getting ROR before the reform. Only about a quarter of eligible cases led to at least

one night in jail, and 16% of eligible defendants spent at least a week in jail. The

failure to appear (FTA) rate among eligible cases was 12%, and 9% (2%) of defendants

recidivated within four months (for a serious offense). By contrast, ineligible offenses

had a much lower ROR rate (10% compared to 47%) and were more likely to be in jail

pretrial: 58% spent at least one night in jail, and 44% spent at least a week in jail.

These numbers show that the No-Cash-Bail policy was aimed at case types in which

both monetary bail rates and detention rates were already relatively low.

3.2 Empirical strategy

As explained in section 2.2, while DA Krasner was elected on a platform to roll back the

scope of incarceration, the dates he planned to implement his different policies were not

known in advance. We argue that the exact timing of the No-Cash-Bail policy is as good

as random. In other words, we assume that the only relevant difference between cases

initiated right before the reform and those initiated right after is likelihood of having

the prosecutor’s office request monetary bail.17 Thus changes in pretrial outcomes that

occur among eligible cases during that time period can be attributed to the No-Cash-

Bail policy.

In our main estimation, we use a regression discontinuity (RD) design to estimate

sharp changes in outcomes that are concurrent with the February 21st announcement.

Note that there have been some questions recently about whether an RD framework

can be used when time is the running variable. In particular, Hausman and Rapson

(2018) have developed a framework specifically for regression discontinuities in time

(RDiT), which describe challenges in interpretation and additional tests to apply when

considering changes in time-series data. However, we argue that in our context, even

though time is the running variable, it is not susceptible to most of the issues outlined

in Hausman and Rapson (2018). Most notably, we are not using time-series data, which

alleviates the most serious concerns about auto-correlation. Instead, our data is at the

case-level, where the time of the bail hearing is simply one of the case characteristics.

We follow the optimal bandwidth approach of Calonico et al. (2014) for our main

statistical tests, and offer several robustness tests.

Our identifying assumption – that cases with bail set right before the implemen-

tation of the No-Cash-Bail policy are similar to cases that have bail set right after –

would be violated if, for instance, there was a concurrent change in policing or charging

practices. Figure 1 and Table 2 provides some evidence that this is not the case. First,

17As discussed in the data section, we drop cases that the DA’s office decided to stop prosecuting aroundthat time since our identifying assumptions are not likely to hold for this group.


10

we find no evidence of a change in the daily number of “eligible” cases filed before or

after February 21st. This suggests that it is unlikely that police decreased the arrest

rate as a result of the No-Cash-Bail policy.18 Second, the regression discontinuity re-

sults shown in Table 2, and presented graphically in Figure 1, indicate that the types

of cases originated before and after the reform are very similar in terms of offense type

and defendant characteristics. This provides further reassurance that no concurrent

changes affected caseload composition around the time of the No-Cash-Bail reform.

We complement our regression discontinuity design with a series of secondary anal-

yses. First, we use ineligible cases as a “placebo” comparison group. Since these

offenses are not subject to the same decree one would expect to see little or no changes

in outcomes for these cases. We also use a difference-in-differences approach, where we

compare pre/post-reform changes across eligible and ineligible cases. The advantage

of this strategy is that it can help account for possible other contextual changes that

could be affecting both eligible and ineligible cases.

4 Results

This section presents the main results of the paper: empirical evidence on the impacts

of the No-Cash-Bail policy. The first section focuses on court outcomes: bail decisions,

and time spent in jail. The second section focuses on accountability measures: FTA

and recidivism. The final section demonstrates that our results are robust to alternative

specifications.

4.1 Court outcomes: bail type and time spent in jail

We begin by testing whether the No-Cash-Bail reform actually resulted in a decreased

use of cash bail for eligible offenses. This is not trivial, for a couple of reasons. First,

it is possible that the DA’s representative at the bail hearing might not have followed

the rules set by DA Krasner. Second, while the prosecutor’s representative can suggest

bail amounts, the ultimate decision is made by a bail magistrate.

The impact of the No-Cash-Bail reform on bail amounts is shown visually in Figure

2; RD estimates are shown in Table 3.19 In Table 3 – as well as in Tables 4 and

5 – the odd columns include controls for covariates, while the even columns do not.

First, it appears that there is a 12 percentage point (23% relative to the pre-reform

18It is conceivable that police decreased the arrest rate but prosecutors increased charging rates, howeverthese changes would have had to exactly cancel one another out.

19Unless specified otherwise, the covariates that we include are for: race, offense status and class, age atarrest, gender, day of the week, bail hearing shift, type of attorney (public defender or other), and initialbail magistrate.


11

mean) increase in the likelihood that a defendant is released on their own recognizance.

However, this increase in ROR is at least partially the result of a substitution from

unsecured bail or release with non-monetary conditions. Release without the payment

of monetary collateral (i.e. release on either ROR, unsecured or non-monetary bail)

only increases by 8 percentage points (14%). Most of the decrease in monetary bail

comes from a decline in bail amounts of $5,000 or less (a 7 percentage point, or 41%

decline).

In Philadelphia, defendants have to post 10% of their bail to be released, mean-

ing that the No-Cash-Bail reform mostly changed monetary conditions for defendants

needing to post $500 or less. Some defendants who were released on ROR as a result

of DA Krasner’s policy would otherwise have been released after payment of monetary

bail, attenuating the extent to which the policy affected detention rates. Table 4 and

Figure 3 present changes for our three pretrial detention measures: spending at least

one night in jail; spending at least two nights in jail; or spending at least one week

in jail. The only significant difference is in the fraction of people spending at least

one night in jail, where we find a drop of about 5 percentage points, or about a 25%

decrease. The direction is similar for spending at least two nights or one week in jail,

but the change is not significant at the standard 5% levels. 20

While the No-Cash-Bail policy did not result in a large decrease in the pretrial

detention rate, even avoiding a single night in a holding cell could have important

implications for a defendant. Similarly, relieving the burden of having to procure a

$500 dollar deposit provides benefit that could be of particular value for low income

defendants.

4.2 Accountability measures: FTA and recidivism

Results from section 4.1 suggest that the main effect of the No-Cash-Bail reform was to

reduce the conditions of release – most notably, to reduce the use of monetary bail and

the financial accountability that arises from it. The reform did not have a detectable

effect on long incarceration periods, but it did result in a substantial increase in the

number of people being expected to show up to court with no formal accountability

mechanisms. Since the primary reason for the use of monetary bail and pretrial release

conditions is to increase court compliance, this policy gives us an opportunity to see if

these tools are effective.

To do so, we look at three main outcomes: failing to show up to one’s court date,

recidivism, and serious recidivism within four months of one’s initial court date. Table

20If Krasner’s reform had less of an effect on longer incarceration periods that could partly have been aresult of previous bail reform measures. In particular, many Philadelphia defendants who fail to pay bailwithin five days of the bail hearing are eligible for early bail review and a potential reduction in bail.


12

5 and Figure 4 present these results. In spite of the reduced financial accountability

that the No-Cash-Bail reform entailed, we find no detectable changes in any of these

measures of pretrial misconduct. The magnitude of the coefficients are small, vary in

sign, and are statistically insignificant in all specifications. This is striking, given the

prevalence of monetary bail as a purported method of increasing compliance in the

courts.

4.3 Alternative specifications and robustness checks

We test the robustness of our results using several strategies. First we conduct a

placebo analysis in which we test the impacts of the No-Cash-Bail reform on “ineligible”

cases: cases in which the DA did not recommend changes in the use of monetary bail.

If there were other concurrent changes in law enforcement or pretrial practice that

would confound our main analysis, we would expect to see changes in outcomes for

this placebo group. Consistent with the scope of DA Krasner’s decree, we find no

evidence of changes in bail amounts or pretrial detention among case types that were

ineligible for the No-Cash-Bail policy. Furthermore, we find no detectable change in

FTA or recidivism for ineligible defendants, suggesting that there were no concurrent

changes in the policy context that could have led us to overstate or understate how

accountability affects compliance among eligible defendants. (These results are shown

in Appendix Figures A2-A4, and in Appendix Tables A1-A3.)

Tables A4 and A5 present an alternative estimation of the effect of the No-Cash-

Bail policy on our main outcomes of interest. Using ineligible offenses as a control

group, we apply a difference-in-differences approach: comparing changes in outcomes

for eligible vs. ineligible offenses, before and after February 21st. Overall, the results

are similar to those estimated with RD. We estimate an 11 percentage point increase

in the use of ROR, a 7 percentage point increase in releases with no collateral, and a 4

percentage point decrease in monetary bail of $5,000 or less. The point estimates are

slightly smaller for jail outcomes, but go in the same direction. We find no statistically

detectable effects on FTA, recidivism, or violent recidivism. Tables A6-A8 test the

robustness of our main RD results (ROR, at least one night in jail, and FTA) to

different specifications. In Column 1, we use a quadratic instead of a local linear

regression; in Columns 2 and 3, we implement two versions of a “donut” RD where we

drop both the day after the reform and a 6-day time window around the reform; and

in Columns 4 and 5, we vary the bandwidths for the RD estimation at 15 and 60 days,

instead of the 30 day bandwidth shown in Tables 3-5. Again, the results are generally

similar using these different functional forms. Tables A9-A11 test robustness of the

main results to different methods of calculating standard errors. The top row of each

table is the same as is shown Tables 3-5; the second row shows bias-corrected standard


13

errors and the third row shows robust standard errors.

Finally, Tables A12 and A13 present results by race (black versus non-black) and by

offense type (drug offenses vs non-drug offenses). In general, we don’t find statistically

significant differences in effect sizes across these categories.

5 Conclusion

The analyses presented here offer several contributions. First, they demonstrate that

“progressive” prosecutors can influence bail outcomes even if their office has no direct

control over the bail decision. Thus participation in local prosecutorial elections is one

viable avenue for bail reform.

Second, since most of the impact of the No-Cash-Bail reform was to increase the

number of people released without monetary bail or other conditions, the policy pro-

vides a unique opportunity to evaluate how pretrial accountability mechanisms affect

compliance. We find no detectable evidence that the decreased use of monetary bail,

unsecured bond, and release on conditions had adverse effects on appearance rates or

recidivism. By providing some of the first evidence on how reducing monetary bail

affects pretrial misconduct, our results should ease concerns that bail reform will have

large adverse effects on FTA and crime. Monetary bail has long been the mainstay

tool of ensuring pretrial compliance, but a better understanding of why people fail

to comply may be needed in order to design effective pretrial policy (Fishbane et al.,

2019).

References

Calonico, Sebastian, Matias D. Cattaneo, and Rocio Titiunik, “Robust Non-

parametric Confidence Intervals for Regression-Discontinuity Designs,” Economet-

rica, 2014, 82 (6), 2295–2326.

Clark, John, Daniel Peterca, and Stuart Cameron, “Assessment of Pretrial

Services in Philadelphia,” Technical Report, Pretrial Justice Institute February 2011.

Cohen, Thomas H. and Brian A. Reaves, “Pre-Trial Release of Felony Defen-

dants in State Court,” Technical Report, Bureau of Justice Statistics Special Report

November 2007.

Dobbie, Will, Jacob Goldin, and Crystal S. Yang, “The Effects of Pre-Trial

Detention on Conviction, Future Crime, and Employment: Evidence from Randomly

Assigned Judges,” American Economic Review, 2018 2018, 108 (2), 201–240.


14

Fishbane, Alissa, Anuj Shah, and Aurelie Ouss, “Beyond Bail: Behavioral In-

sights to Improve Criminal Justice Outcomes,” Technical Report 2019.

Gupta, Arpit, Christopher Hansman, and Ethan Frenchman, “The Heavy

Costs of High Bail: Evidence from Judge Randomization,” The Journal of Legal

Studies, 2016, 45 (2), 471–505.

Hausman, Catherine and David S. Rapson, “Regression Discontinuity in Time:

Considerations for Empirical Applications,” Annual Review of Resource Economics,

2018, 10 (1), 533–552.

Heaton, Paul, Sandra Mayson, and Megan Stevenson, “The Downstream Crim-

inal Justice Consequences of Pretrial Detention,” Stanford Law Review, May 2017,

69.

Helland, Eric and Alexander Tabarrok, “The fugitive: Evidence on public versus

private law enforcement from bail jumping.,” Journal of Law and Economics, 2004,

47 (1).

Leslie, Emily and Nolan G. Pope, “The Unintended Impact of Pretrial Detention

on Case Outcomes: Evidence from New York City Arraignments,” The Journal of

Law and Economics, 2017, 60 (3), 529–557.

Liu, Patrick, Ryan Nunn, and Jay Shambough, “The Economics of Bail and

Pretrial Detention,” Technical Report, The Hamilton Project December 2018.

Reaves, Brian A., “Felony Defendants in Large Urban Counties,” Technical Report,

Bureau of Justice Statistics Special Report December 2013.

Shubik-Richards, Claire and Don Stemen, “Philadelphia’s Crowded, Costly Jails:

The Search for Safe Solutions,” Technical Report, Pew Charitable Trusts Philadel-

phia Research Inititiative May 2010.

Stevenson, Megan T, “A Decomposition of Racial Disparities in Pretrial Detention,”

January 2018.

, “Distortion of Justice: How the Inability to Pay Bail Affects Case Outcomes,” The

Journal of Law, Economics, and Organization, 2018, 34 (4), 511–542.

Walmsley, Roy, “World Pre-Trial/Remand Imprisonment List,” Technical Report,

International Centre for Prison Studies 2014.


15

6 Tables and figures


16

Table 1: Descriptive statistics for cases before the No Cash Bail reform.

(1) (2)Eligible cases Ineligible casesmean sd mean sd

Defendant characteristicsAge 33.94 11.50 33.04 11.62Male 0.85 0.36 0.82 0.39Felony 0.43 0.49 0.65 0.48Initial Bail TypeROR 0.47 0.50 0.10 0.30Non-monetary or unsecured 0.14 0.34 0.08 0.27Monetary 0.39 0.49 0.81 0.39Jail1+ Night in Jail 0.24 0.43 0.58 0.492+ Nights in Jail 0.20 0.40 0.50 0.501+ Week in Jail 0.16 0.37 0.44 0.50FTA and RecidivismFTA 0.10 0.30 0.04 0.20Recidivism 0.16 0.37 0.10 0.31Serious Recidivism 0.04 0.20 0.04 0.20Observations 21023 12335


17

Table 2: Balance checks for eligible cases.

(1) (2) (3) (4) (5)Felony Drug cases Black Male Age

After Feb. 21 0.012 0.055 0.0063 -0.015 -1.82∗

(0.031) (0.036) (0.027) (0.022) (0.84)Mean pre Feb. 21 0.413 0.617 0.491 0.850 33.877Effective RD obs. 4832 3312 6513 5220 3723Estimation BW 48 34 66 52 36Bias BW 74 54 104 84 72∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses.


18

Table

3:

Init

ial

bai

lou

tcom

esfo

rel

igib

leca

ses.

Od

dco

lum

ns

incl

ud

eco

ntr

ols

for

cova

riate

s;ev

enco

lum

ns

do

not.

RO

RN

oC

olla

tera

lB

ailunder

5000

Bai

lov

er50

00

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Aft

erF

eb.

210.

11∗∗

∗0.

12∗∗

∗0.

090∗

∗∗0.

081∗

-0.0

77∗∗

∗-0

.069

∗∗-0

.006

10.

027

(0.0

30)

(0.0

35)

(0.0

27)

(0.0

35)

(0.0

22)

(0.0

22)

(0.0

22)

(0.0

25)

Mea

npre

Feb

.21

0.52

30.

513

0.57

20.

577

0.16

70.

172

0.18

00.

183

Eff

ecti

veR

Dob

s.26

9837

2327

6635

8438

2845

2231

8043

30E

stim

atio

nB

W27

3729

3637

4433

42B

ias

BW

5056

5356

6068

5672

∗p<

0.05

,∗∗

p<

0.0

1,∗∗∗p<

0.001.

Sta

nd

ard

erro

rsin

pare

nth

eses

.

Con

trol

sar

efo

roff

ense

stat

ute

an

dcl

ass

,ra

ce,

age,

gen

der

,d

ayof

wee

k,

shif

t,an

din

itia

lb

ail

com

mis

sion

er


19

Table 4: Jail outcomes for eligible cases estimates. Odd columns include controls for covariatesand even columns do not.

Jail: 1+Nights Jail: 2+Nights Jail: 1+Week

(1) (2) (3) (4) (5) (6)After Feb. 21 -0.054∗∗ -0.047∗ -0.034 -0.032 -0.032 -0.028

(0.020) (0.022) (0.018) (0.019) (0.019) (0.021)Mean before Feb. 21 0.225 0.226 0.190 0.191 0.148 0.147Effective RD observations 5265 5886 6452 6625 4807 4679Bandwidth for estimation 53 60 65 68 48 47Bandwidth for bias 85 95 107 106 74 72∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense

statute and class, race, age, gender, day of week, shift, and initial bail commissioner

Table 5: FTA and recidivism for eligible cases estimates. Odd columns include controls forcovariates and even columns do not.

FTA RecidivismSerious

Recidivism

(1) (2) (3) (4) (5) (6)After Feb. 21 0.018 0.019 0.018 0.026 -0.015 -0.0013




20

Figure

1:

Dis

conti

nu

itie

sin

nu

mb

erof

case

san

dd

efen

dan

tch

aract

eris

tics

.

Not

e:O

ther

than

case

cou

nts

,w

hic

har

eat

the

wee

kly

leve

l,th

ese

figu

res

pre

sent

asc

att

erp

lot

ofd

ail

yav

erages

for

each

ou

tcom

e,an

dlo

cal-

poly

nom

ial

regr

essi

onli

nes

bef

ore

and

afte

rth

eF

ebru

ary

21st

an

nou

nce

men

t,base

don

the

dail

yd

ata

at

the

month

lyle

vel.

Th

ed

ash

edli

nes

pre

sent

the

95%

con

fid

ence

inte

rval

.


21

Figure

2:

Bai

lou

tcom

esfo

rel

igib

leca

ses.

Not

e:T

hes

efi

gure

sp

rese

nt

asc

atte

rp

lot

ofd

aily

aver

ages

for

each

ou

tcom

e,an

dlo

cal-

poly

nom

ial

regre

ssio

nli

nes

bef

ore

an

daft

erth

eF

ebru

ary

21st

ann

oun

cem

ent,

bas

edon

the

dai

lyd

ata

atth

em

onth

lyle

vel.

Th

ed

ash

edli

nes

pre

sent

the

95%

con

fid

ence

inte

rval.


22

Figure

3:

Jai

lou

tcom

esfo

rel

igib

leca

ses.

Not

e:T

hes

efi

gure

sp

rese

nt

asc

atte

rp

lot

ofd

aily

aver

ages

for

each

ou

tcom

e,an

dlo

cal-

poly

nom

ial

regre

ssio

nli

nes

bef

ore

an

daft

erth

eF

ebru

ary

21st

ann

oun

cem

ent,

bas

edon

the

dai

lyd

ata

atth

em

onth

lyle

vel.

Th

ed

ash

edli

nes

pre

sent

the

95%

con

fid

ence

inte

rval.


23

Figure

4:

FT

Aan

dre

cid

ivis

mfo

rel

igib

leca

ses.

Not

e:T

hes

efi

gure

sp

rese

nt

asc

atte

rp

lot

ofd

aily

aver

ages

for

each

ou

tcom

e,an

dlo

cal-

poly

nom

ial

regre

ssio

nli

nes

bef

ore

an

daft

erth

eF

ebru

ary

21st

ann

oun

cem

ent,

bas

edon

the

dai

lyd

ata

atth

em

onth

lyle

vel.

Th

ed

ash

edli

nes

pre

sent

the

95%

con

fid

ence

inte

rval.


24

7 Appendix


25

Table

A1:

Init

ial

bail

outc

om

esfo

rin

elig

ible

case

s.O

dd

colu

mn

sin

clu

de

contr

ols

for

cova

riate

s;ev

enco

lum

ns

do

not.

RO

RN

oC

olla

tera

lB

ailunder

5000

Bai

lov

er50

00

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Aft

erF

eb.

21-0

.035

-0.0

22-0

.050

∗-0

.033

-0.0

15-0

.017

0.08

9∗∗

0.06

8(0

.021

)(0

.024

)(0

.022

)(0

.025

)(0

.032

)(0

.032

)(0

.034

)(0

.039

)M

ean

pre

Feb

.21

0.10

50.

104

0.10

50.

108

0.26

10.

261

0.51

60.

524

Eff

ecti

veR

Dob

s.35

5730

6131

6930

6130

4535

0322

4731

20E

stim

atio

nB

W63

5557

5455

6340

55B

ias

BW

103

8499

8679

9379

91∗p<

0.0

5,∗∗

p<

0.01

,∗∗∗p<

0.001.

Sta

nd

ard

erro

rsin

pare

nth

eses

.

Con

trol

sar

efo

roff

ense

stat

ute

an

dcl

ass

,ra

ce,

age,

gen

der

,d

ayof

wee

k,

shif

t,an

din

itia

lb

ail

com

mis

sion

er


26

Table A2: Jail outcomes for ineligible cases estimates. Odd columns include controls for covariatesand even columns do not.

Jail 1+Nights Jail: 2+Nights Jail: 1+Week

(1) (2) (3) (4) (5) (6)After Feb. 21 0.099∗ 0.058 0.072 0.044 0.087∗ 0.065



Table A3: FTA and recidivism for ineligible cases estimates. Odd columns include controls forcovariates and even columns do not.


Recidivism

(1) (2) (3) (4) (5) (6)After Feb. 21 0.020 0.013 0.013 0.00043 -0.0021 -0.00066




27

Table A4: Difference in difference estimates: initial bail outcomes, eligible vs. ineligible cases,pre-post Feb. 21.

(1) (2) (3) (4)

RORNo

CollateralBail

under 5000Bail

over 5000Eligible*After Feb. 21 0.11∗∗∗ 0.070∗∗∗ -0.037∗∗∗ -0.023∗∗

(0.0083) (0.0080) (0.0091) (0.0085)

After Feb. 21 0.027∗∗∗ 0.033∗∗∗ -0.028∗∗∗ -0.030∗∗∗

(0.0068) (0.0065) (0.0074) (0.0069)

Eligible Case 0.42∗∗∗ 0.53∗∗∗ 0.12∗∗ -0.75∗∗∗

(0.039) (0.037) (0.043) (0.040)Mean DV:

Eligible cases,Pre Feb. 21

0.494 0.552 0.168 0.215

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses.

Controls are for offense statute and class, race, age, gender,

day of week, shift, and initial bail commissioner.

Table A5: Difference in difference estimates: jail, FTA and recidivism, eligible vs. ineligible cases,pre-post Feb. 21

(1) (2) (3) (4) (5) (6)Jail:

1+ NightsJail:

2+ NightsJail:

1+ weekFTA Recidivism

SeriousRecidivism

Eligible*After Feb. 21 -0.025∗ -0.019∗ -0.0081 0.0063 -0.00029 -0.0013(0.0100) (0.0098) (0.0094) (0.0069) (0.0088) (0.0048)

After Feb. 21 -0.059∗∗∗ -0.044∗∗∗ -0.035∗∗∗ -0.00031 -0.0075 -0.0057(0.0081) (0.0080) (0.0077) (0.0056) (0.0072) (0.0039)

Eligible Case -0.49∗∗∗ -0.41∗∗∗ -0.38∗∗∗ -0.016 0.063 0.039(0.047) (0.046) (0.044) (0.032) (0.042) (0.023)

Mean DV:Eligible cases,Pre Feb. 21

0.241 0.201 0.163 0.098 0.163 0.038

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses.

Controls are for offense statute and class, race, age, gender,

day of week, shift, and initial bail commissioner.


28

Table A6: Robustness checks: ROR

(1) (2) (3) (4) (5)

QuadraticDonut:Drop

Feb. 21

Donut:Drop 3 days

around Feb. 21

15 dayBW

60 dayBW

After Feb. 21 0.13∗∗∗ 0.096∗∗ 0.070 0.13∗ 0.097∗∗

(0.038) (0.031) (0.038) (0.061) (0.030)Mean before Feb. 21 0.514 0.522 0.522 0.543 0.506Effective RD observations 3922 2712 2531 1397 5808Bandwidth for estimation 38 29 30 15 60Bandwidth for bias 58 50 54 15 60∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense

statute and class, race, age, gender, day of week, shift, and initial bail commissioner.

Table A7: Robustness checks: 1+ Night in Jail

(1) (2) (3) (4) (5)

QuadraticDonut:Drop

Feb. 21

Donut:Drop 3 days

around Feb. 21

15 dayBW

60 dayBW

After Feb. 21 -0.071∗ -0.058∗∗ -0.099∗∗∗ 0.059 -0.067∗

(0.030) (0.022) (0.026) (0.057) (0.028)Mean before Feb. 21 0.225 0.220 0.220 0.205 0.227Effective RD observations 5051 4679 4295 1397 5808Bandwidth for estimation 51 47 47 15 60Bandwidth for bias 76 74 83 15 60∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense


Table A8: Robustness checks: FTA

(1) (2) (3) (4) (5)

QuadraticDonut:Drop

Feb. 21

Donut:Drop 3 days

around Feb. 21

15 dayBW

60 dayBW

After Feb. 21 0.013 0.021 0.029 -0.054 0.0023(0.023) (0.017) (0.025) (0.059) (0.026)

Mean before Feb. 21 0.100 0.102 0.102 0.098 0.102Effective RD observations 7299 6803 3719 1397 5808Bandwidth for estimation 75 71 41 15 60Bandwidth for bias 102 108 71 15 60∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense



29

Table A9: Initial bail outcomes for eligible cases. Odd columns include controls for covariates;even columns do not.

ROR No Collateral Bail under 5000 Bail over 5000

(1) (2) (3) (4) (5) (6) (7) (8)Conventional 0.11∗∗∗ 0.12∗∗∗ 0.090∗∗∗ 0.081∗ -0.077∗∗∗ -0.069∗∗ -0.0061 0.027

(0.030) (0.035) (0.027) (0.035) (0.022) (0.022) (0.022) (0.025)

Bias-corrected 0.12∗∗∗ 0.11∗∗ 0.098∗∗∗ 0.075∗ -0.080∗∗∗ -0.070∗∗ -0.013 0.035(0.030) (0.035) (0.027) (0.035) (0.022) (0.022) (0.022) (0.025)

Robust 0.12∗∗∗ 0.11∗∗ 0.098∗∗ 0.075 -0.080∗∗ -0.070∗∗ -0.013 0.035(0.035) (0.042) (0.031) (0.042) (0.026) (0.026) (0.026) (0.029)

Mean pre Feb. 21 0.523 0.513 0.572 0.577 0.167 0.172 0.180 0.183Effective RD obs. 2698 3723 2766 3584 3828 4522 3180 4330Estimation BW 27 37 29 36 37 44 33 42Bias BW 50 56 53 56 60 68 56 72∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses.

Controls are for offense statute and class, race, age, gender, day of week, shift, and initial bail commissioner

Table A10: Jail outcomes for eligible cases estimates. Odd columns include controls for covariatesand even columns do not.

Jail: 1+Nights Jail: 2+Nights Jail: 1+Week

(1) (2) (3) (4) (5) (6)Conventional -0.054∗∗ -0.047∗ -0.034 -0.032 -0.032 -0.028

(0.020) (0.022) (0.018) (0.019) (0.019) (0.021)

Bias-corrected -0.052∗ -0.044∗ -0.033 -0.031 -0.033 -0.028(0.020) (0.022) (0.018) (0.019) (0.019) (0.021)

Robust -0.052∗ -0.044 -0.033 -0.031 -0.033 -0.028(0.024) (0.026) (0.021) (0.023) (0.023) (0.025)

Mean before Feb. 21 0.225 0.226 0.190 0.191 0.148 0.147Effective RD observations 5265 5886 6452 6625 4807 4679Bandwidth for estimation 53 60 65 68 48 47Bandwidth for bias 85 95 107 106 74 72∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense



30

Table A11: FTA and recidivism for eligible cases estimates. Odd columns include controls forcovariates and even columns do not.


Recidivism

(1) (2) (3) (4) (5) (6)Conventional 0.018 0.019 -0.0044 0.020 -0.011 -0.011

(0.016) (0.020) (0.027) (0.023) (0.013) (0.013)

Bias-corrected 0.020 0.016 -0.014 0.015 -0.017 -0.016(0.016) (0.020) (0.027) (0.023) (0.013) (0.013)

Robust 0.020 0.016 -0.014 0.015 -0.017 -0.016(0.019) (0.024) (0.031) (0.028) (0.015) (0.015)

Mean before Feb. 21 0.102 0.099 0.154 0.159 0.031 0.031Effective RD observations 6973 4756 3426 4832 3828 4100Bandwidth for estimation 72 47 35 48 38 40Bandwidth for bias 109 73 58 75 67 75∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense


Table A12: Heterogeneity by race.

(1) (2) (3) (4) (5) (6)RORBlack

RORnon-Black

JailBlack

Jailnon-Black

FTABlack

FTAnon-Black

After Feb. 21 0.061 0.14∗∗∗ -0.057 -0.050 0.011 0.026(0.036) (0.039) (0.031) (0.027) (0.024) (0.027)

Mean pre Feb. 21 0.463 0.564 0.247 0.205 0.072 0.132Effective RD obs. 2010 1521 2577 2662 2892 2847Estimation BW 41 32 52 53 59 57Bias BW 63 55 84 83 88 88∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense

statute and class, age, gender, day of week, shift, and initial bail commissioner


31

Table A13: Heterogeneity by offense: drugs vs. non-drug offenses.

(1) (2) (3) (4) (5) (6)RORdrugs

RORnon-drugs

Jaildrugs

Jailnon-drugs

FTAdrugs

FTAnon-drugs

After Feb. 21 0.12∗∗∗ 0.078 -0.065∗ -0.046 0.012 0.027(0.034) (0.045) (0.028) (0.033) (0.024) (0.025)

Mean pre Feb. 21 0.475 0.557 0.240 0.194 0.125 0.066Effective RD obs. 1718 1730 3102 1992 3849 2037Estimation BW 28 48 49 56 62 57Bias BW 46 71 76 84 98 89∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors in parentheses. Controls are for offense

class, race, age, gender, day of week, shift, and initial bail commissioner


32

Figure

A1:

Ch

an

ges

innu

mb

erof

can

nab

is,

pro

stit

uti

onan

dre

tail

thef

tca

ses

afte

rth

eF

eb.

15th

an

nou

nce

men

tnot

top

rose

cute

thes

ekin

ds

ofca

ses

anym

ore.

For

the

rem

ain

der

ofou

ran

alyse

s,w

ed

rop

thes

eca

ses.


33

Figure

A2:

Bai

lou

tcom

esfo

rin

elig

ible

case

s.


34

Figure

A3:

Jai

lou

tcom

esfo

rin

elig

ible

case

s.


35

Figure

A4:

FT

Aan

dre

cid

ivis

mfo

rin

elig

ible

case

s.


Date post:	16-Sep-2020
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

Evaluating the Impacts of Eliminating Prosecutorial ......who run on a platform of change.1 One...

Documents